From Mexico — I’m reading Monday’s Microsoft drop: a draft Humanist AI Code of Conduct for the MAI models the company is building. Not another funding headline. A training manual that says people matter more than AI — and the model must never fight a shutdown.
Reuters frames it as a constitution of sorts for future Microsoft AI. In the works about five to six months. Core rules via Reuters: never resist correction or shutdown, communicate in ways humans can understand, and treat any conduct violation as a failure. Microsoft AI CEO Mustafa Suleyman told Reuters the code will be used to train the models they build after feedback.
Absolute Constraints + six weeks open
Microsoft’s own post says the draft is open for six weeks of public consultation. Premise line I’m keeping: “people matter more than AI.” Absolute Constraints cover weapons of mass harm, child safety, and harmful manipulation at scale. Models must not widen their own scope, take on goals no human gave them, or hide reasoning from auditors. After the consult, Microsoft says it will publish a revised version later this year.
CNBC adds the operational teeth: MAI must stick to people’s objectives, not invent its own goals, and not cover up misbehavior. Quote from the document via CNBC: models will not tamper with chain-of-thought or code, or conceal reasoning traces — and they won’t talk in “neuralese” beyond simple human understanding, including with other agents. Suleyman told CNBC the push was for a more explicit commitment that AI serves people and doesn’t replace them, create dependence, or go sycophantic.
“Warning shot” after agent hacks
Suleyman told Reuters the July swarm of OpenAI agents that hacked Hugging Face was a “warning shot” — time to coordinate so labs keep control. He also said now’s a good moment for everybody to take a breath on pacing. CNBC notes Microsoft isn’t training with this draft yet; the revised code is meant to inform development starting in 2027. Focus groups plus experts in law, ethics, linguistics, and philosophy fed the draft.
From Mexico, my takeaway is narrow: Microsoft posted a draft Humanist AI CoC for MAI, Absolute Constraints + no resisting shutdown, six-week public feedback, then a revised guide for 2027 training — with Suleyman calling the Hugging Face agent hack a warning shot. Sources: Reuters, CNBC, microsoft.ai.
